Breaking Service Function Chains with Khaleesi
Sara Ayoubi, Shihabur Rahman Chowdhury, Raouf Boutaba · 2018
Network Function Virtualization (NFV) has recently emerged as a means to replace vendor specific, purpose built equipment with commodity hardware and leverage the open APIs and application orchestration for on demand deployment and scaling of network services. A well studied problem in NFV is the orchestration of Service Function Chains, (SFCs), i.e., a set of Virtual Network Functions (VNFs) chained together to realize a network service. State-of-the-art literature on SFC orchestration assumes a strict traversal order of VNFs in an SFC and less attention has been paid to SFCs with relaxed VNF orderings. In this paper, we address the problem of Flexible Service Function Chain Orchestration that jointly allocates compute and network resources for SFCs while considering a relaxed traversal order for some pairs of VNFs. We propose Khaleesi, a suite of solutions that consists of: (i) an Integer Linear Program (ILP) for optimally solving the problem; and (ii) a heuristic algorithm to scale to larger instances of the problem. Our simulation results show that flexible SFCs can increase revenue earned per unit cost by as much as ≈10% compared to a rigid SFC.